Machine Learning-Enabled Vertically Integrated Heterogeneous Manycore Systems for Big-Data Analytics

Download or read book Machine Learning-Enabled Vertically Integrated Heterogeneous Manycore Systems for Big-Data Analytics written by Biresh Kumar Joardar and published by . This book was released on 2020 with total page 101 pages. Available in PDF, EPUB and Kindle. Book excerpt: The rising use of deep learning and other big-data algorithms has led to an increasing demand for hardware platforms that are computationally powerful, yet energy-efficient. Heterogeneous manycore architectures that integrate multiple types of cores on a single chip present a promising direction in this regard. However, designing these new architectures often involves optimizing multiple conflicting objectives (e.g., performance, power, thermal, reliability, etc.) due to the presence of a mix of computing elements and communication methodologies; each with a different requirement for high-performance. This has made the design, and evaluation of new architectures an increasingly challenging problem. Machine Learning algorithms are a promising solution to this problem and should be investigated further. This dissertation focuses on the design of high-performance and energy efficient architectures for big-data applications, enabled by data-driven machine learning algorithms. As an example, we consider heterogeneous manycore architectures with CPUs, GPUs, and Resistive Random-Access Memory (ReRAMs) as the choice of hardware platform in this work. The disparate nature of these processing elements introduces conflicting design requirements that need to be satisfied simultaneously. In addition, novel design techniques like Processing-in-memory and 3D integration introduces additional design constraints (like temperature, noise, etc.) that need to be considered in the design process. Moreover, the on-chip traffic pattern exhibited by different big-data applications (like many-to-few-to-many in CPU/GPU-based manycore architectures) need to be incorporated in the design process for optimal power-performance trade-off. However, optimizing all these objectives simultaneously leads to an exponential increase in the design space of possible architectures. Existing optimization algorithms do not scale well to such large design spaces and often require more time to reach a good solution. In this work, we highlight the efficacy of machine learning algorithms for efficiently designing a suitable heterogeneous manycore architecture. For large design space exploration problems, the proposed machine learning algorithm can find good solutions in significantly less amount of time than exiting state-of-the-art counterparts.On overall, this work focuses on the design challenges of high-performance and energy efficient architectures for big-data applications, and proposes machine learning algorithms capable of addressing these challenges.